Simple-LJX

Theo Moore-Calters, Vijay,
{company}
Code

Simple-LJX Overview

Simple-LJX is a collection of locomotion-based MJX environments designed for end-to-end training using reinforcement learning.

  • Built for simplicity
  • Powered by Brax (a well-respected JAX-based RL training framework)
  • Each environment can be viewed live online using our {company} links

Our Research Mels

Quadruped Model

Humanoid Model

Humanoid Model

Quadruped Model

  • Actuators: 17
  • Degrees of Freedom: 24
  • Locomotion observation shape: 276
    • Joint positions: 22
    • Joint velocities: 23
    • Inertia on each body: 170
    • Global velocities on each body: 78
    • Actuator forces: 23
  • Actuators: 12
  • Degrees of Freedom: 15
  • Locomotion observation shape: 27
    • Joint positions: 13
    • Joint velocities: 14
We have a catalogue of environments for you to try out, from sprints to longjump .... further build up

Locomotion

teaser

Simulation Environment

Fetch policy by Quadruped

Fetch policy by Humanoid

Hurdle policy by Humanoid

Gaps policy by Humanoid

Long Jump policy by Humanoid

Long Jump policy by Quadruped

Sprint policy by Humanoid

Sprint policy by Quadruped